1 / 31

Biologically Inspired Computing: Operators for Evolutionary Algorithms

Biologically Inspired Computing: Operators for Evolutionary Algorithms. Some basic operators. How to think of genetic operators. Selection represents our strategy for deciding which areas of the search space to focus our efforts on.

alexia
Télécharger la présentation

Biologically Inspired Computing: Operators for Evolutionary Algorithms

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Biologically Inspired Computing: Operators for Evolutionary Algorithms Some basic operators

  2. How to think of genetic operators Selection represents our strategy for deciding which areas of the search space to focus our efforts on. Operators provide our means of generating new candidate solutions. • We want operators to have a fair chance of yielding good new solutions (small change, and/or combine bits from solutions we already know are good) • We also (obviously) want to be able to potentially explore the whole space.

  3. Genetic Operators / Variation Methods Often we have used a k-ary encoding, in which a candidate solution is just a list of L numbers, each of which can be anything from 0 to k-1 (or 1 to k). E.g. our simple bin-packing representation uses a k-ary encoding. So, for 100 items and 5 bins, this would be a 5-ary encoding of length L=100. These might be candidate solutions from a k-ary encoding with L = 10 and k = 20: [17, 2, 19, 1, 1, 1, 5, 11, 12, 2] [16, 19, 2, 19, 2, 3, 4, 7, 5, 2]

  4. Mutation in k-ary encodings Single-gene new-allele mutation: Choose a gene at random, and change it to a random new value. E.g. 352872  312872 M-random-gene new-allele mutation: (Isn’t it obvious?) Repeat the above M times. For example, in a 1000-gene problem, we might use 5-gene mutation. This will result in anything from 1 to 5 new gene values.

  5. Mutation in k-ary encodings Single-gene random-allele mutation: Choose a gene at random, and change it to a random value. This is the same as single-gene new-allele mutation, except that it doesn’t take care to make sure we have a new value for the gene. So, often (especially if k is small) it will lead to no change at all. But that’s not a problem – in the EA context, it means that the next generation contains an extra copy of an individual that survived selection (so is probably quite good), and in fact it might not be in the new population otherwise. M-distinct-gene new-allele mutation: Same as M-random-gene, but making sure that Mdifferent genes are changed.

  6. Genewise mutation with strength m; Go through the solution gene by gene, and mutate every one of them with probability m. E.g. if L = 100 we might use m = 0.01 – usually, just one gene will get changed, but possibly none, and possibly 2 or more. There is even a tiny chance of all of them being changed. Single-gene mutation: Choose a gene at random, and add a small random deviation to it. Often chosen from a Gaussian distribution. Vector mutation: Generate a small random vector of length L, and add it to the whole thing.

  7. Crossover in k-ary encodings One-Point Crossover. For encodings of length L, one point crossover works as follows: • Choose a crossover point k randomly (a number from 1 to L-1) • The first k genes of the child will be the same as the first k genes of parent 1 • The genes of the child from gene k+1 onwards will be the same as those of parent 2.

  8. Crossover in k-ary encodings Two-Point Crossover. For encodings of length L, two point crossover works as follows: • Choose two crossover points j and k randomly, making sure that 1 <= j < k < L • Genes 1 to j of the child will be the same as genes 1 to j of parent 1; • genes j+1 to k of the child will be the same as genes j to k of parent 2. • the remainder of the child will be the same as parent 1.

  9. 1-point example: Parent1: 1, 3, 4, 3, 6, 1, 3, 6, 7, 3, 1, 4 Parent2: 3, 5, 2, 6, 7, 1, 2, 5, 4, 2, 2, 8 Random choice of k = 6 Child: 1, 3, 4, 3, 6, 1, 2, 5, 4, 2, 2, 8 2-point example: Parent1: 1, 3, 4, 3, 6, 1, 3, 6, 7, 3, 1, 4 Parent2: 3, 5, 2, 6, 7, 1, 2, 5, 4, 2, 2, 8 Random choices: j = 3, k = 10 Child: 1, 3, 4,6, 7, 1, 2, 5, 4, 2,1, 4

  10. Crossover in k-ary encodings Uniform Crossover. For encodings of length L, Uniform crossover works as follows: For each gene i from 1 to L Let c = 1 or 2, with equal probability. Make Gene i of the child the same as gene i of Parent c In other words, we simply create the child one gene at a time, flipping a coin each time to decide which parent to take the gene from.

  11. Uniform crossover example: Parent1: 1, 3, 4, 3, 6, 1, 3, 6, 7, 3, 1, 4 Parent2: 3, 5, 2, 6, 7, 1, 2, 5, 4, 2, 2, 8 If our random choices in order were: 121221121221 the child would be: Child: 1, 5, 4, 6, 7, 1, 3, 5, 7, 2, 2, 4

  12. Operators in order-based encodings Often, our chosen encoding for a problem will be a permutation. This is the usual encoding for problems such as the travelling salesperson problem, and many others (as we will see in a later lecture). Clearly, k-ary encoding operators are invalid in this case. The following slides show common operators that can be used.

  13. Mutation in order-based encodings In all examples we will assume our encoding is a permutation of L items, where L = 10. swap mutation is simply this: Choose randomly two distinct genes, i and j, and then swap the genes at these positions. Adjacent-swap mutation is simply this: Choose any gene j randomly, then swap it with gene j+1 (modulo L).

  14. Swap mutation example: Parent: A C J I B D E G H F Random choices: i = 3; j = 7 Child: A C E I B D J G H F Adjacent-swap mutation example Parent: A C J I B D E G H F Random choice: j = 8 Child: A C J I B D E H G F

  15. Mutation in order-based encodings k-Inversion-mutation: Choose a random chunk of the parent of size k. That is, choose a random gene j from 1 to L, and the chunk is the section from genes j to j+k-1(modulo L) inclusive. Reverse the genes in this chunk. (note: when k = 2, what is this identical to?)

  16. Example of k-Inversion mutation Here is 3-inversion mutation applied to: A B D G F C E Choose a random chunk of size 3: A B D G F C E Reverse the genes in this chunk: A G D B F C E

  17. Mutation in order-based encodings Variable-k-Inversion-mutation: Choose k randomly between 2 and L. Then do a k-inversion mutation.

  18. k-inversion examples. 4-inversion mutation: Parent: A C J I B D E G H F Random choice of j = 1 Child: I J C A B D E G H F Variable-k mutation: Parent: A C J I B D E G H F Random choices: k = 7; j = 7 Child: H G E I B D J C A F (make sure you understand this one)

  19. Crossover in order-based encodings There are various sensible ways to produce a valid child that combines aspects of two different permutation parents. What might be best depends very much on the application. We will describe two generic order based operators, but others will come up in later lectures looking at particular applications.

  20. k-geneOrder-based Crossover Works like this: First, make the child a copy of parent 1. Next, randomly choose k distinct genes of the child. Next, reorder the values of these genes so that they match their order in parent 2. Naturally, there is also an operator that we can call: Variable-k-order based crossover. I will leave you to work out what that is.

  21. 4-gene order based crossover example. Parent 1: A C J I B D E G H F Parent 2: F E A I H J D B C G Random choices of 4 distinct genes: 2, 5, 6, 9 Child is initially same as parent 1 – shown here with the random choices highlighted: Child: A C J I B D E G H F The order of these four gene values in Parent 2 is: H, D, B, C So, we impose that ordering on the child, but keeping them in the same positions: Child: A H J I DB E G C F

  22. k-genePosition-based Crossover Works like this: First, make the child a copy of parent 1. Next, randomly choose k distinct gene positions of the child. Let V be the set of gene values at these positions. Next, copy the genes of parent 2 that are not in V into the child, overwriting the child’s other genes, in their parent 2 order. Naturally, there is also an operator that we can call: Variable-k-position based crossover. I will leave you to work out what that is. An example is rather necessary in this case – all should become clear after the next slide.

  23. 4-gene position based crossover example. Parent 1: A C J I B D E G H F Parent 2: F E A I H J D B C G Random choices of 4 distinct positions: 2, 5, 6, 9 Child is initially same as parent 1 – shown here with the random positions highlighted: Child: A C J I B D E G H F Now let’s blank out all of the other genes: Child: * C * * B D * * H * We proceed by filling the child up with its missing genes, in the order they appear in Parent 2: This order is: F E A I J G. So the child becomes: Child: F C E A B D I J H G

  24. Operators for real-valued k-ary encodings Here the chromosome is a string of k real numbers, which each may or may not have a fixed range (e.g. between −5 and 5). e.g. 0.7, 2.8, −1.9, 1.1, 3.4, −4.0, −0.1, −5.0, … All of the mutation operators for k-ary encodings, as previously described in these slides, can be used. But we need to be clear about what it means to randomly change a value. In the previous slides for k-ary mutation operators we assumed that a change to a gene mean to produce a random (new) value anywhere in the range of possible values …

  25. Operators for real-valued k-ary encodings that’s fine … we can do that with a real encoding, but this means we are choosing the new value for a gene uniformly at random.

  26. Mutating a real-valued gene using a uniform distrution Range of allowed values for gene: 0—10 New value can be anywhere in the range, with any number equally likely.

  27. But, in real-valued encodings, we usually use Gaussian (‘Normal’) distributions … so that the new value is more likely than not to be close to the old value. typically we generate a perturbation from a Gaussian distribution, like this one, and add that perturbation to the old value. 0.3 0.2 Probability of choosing this perturbation 0.1 0 −1 −0.5 0 0.5 1 perturbation

  28. Mutation in real-valued encodings Most common is to use the previously indicated mutation operators (e.g. single-gene, genewise) but with Gaussian perturbations rather than uniformly chosen new values.

  29. Crossover in real-valued encodings All of the k-ary crossover operators previously described can be used. But there are other common ones that are only feasible for real-valued encodings:

  30. Box and Line crossover operators for real-valued chromosomes – figure is from this paper: http://cdn.intechopen.com/pdfs/30229/InTech-The_roles_of_crossover_and_mutation_in_real_coded_genetic_algorithms.pdf Treat the parents like vectors Fig assumes you are crossing over two 2-gene parents, x (x1,x2) and y (y1,y2) Box: child is (x1 + u1 (y1 – x1), x2 + u2(y2 – x2) ) Where u1 and u2 are uniform random numbers between 0 and 1 Line: child is: x + u (y – x) or (x1+ u(y1 – x1) , x2 + u (y2 – x2) ) Where u is a random number between 0 and 1

  31. Box and Line crossover: general form Parent1: x1, x2, x3, …., xN Parent2: y1, y2, y3, …, yN given parameter α(typically values are 0, 0.1 or 0.25) General form: Child is ((x1 – α) + u (y1 – x1), ((x2 – α) + u (y2 – x2), … etc) Where: u is a uniform random number between 0 and 1+2α Line crossover: α= 0 ; u is generated once for each crossover, and is the same for every gene. Extended line crossover: α> 0, u is generated once for each crossover, and is the same for every gene. Box crossover: α= 0 ; u is different for every gene Extended box crossover: α> 0 ; u is different for every gene

More Related